Specifying a Climate Bet
As I mentioned in the comments on this post, I am currently in the process of negotiating a bet on Anthropogenic Global Warming (AGW) with another blogger. The challenges are interesting, so I thought I’d give you a peek inside the sausage factory.
The first point to note is that the potential counterparty and I are good candidates for a bet. We both are acting as reasonable Bayesians (neither of us have extreme, ideologically-driven views). We both appear to have a decent grasp of the domain. We both have publicly stated our beliefs.
I believe that the underlying warming signal is +.05 deg C per decade and he thinks it’s +.15 deg C per decade. We have agreed that an over-under bet on a decadal trend of +.10 deg C is fair and that 2000 to 2020 should be the measurement period. So far so good. But now things get sticky.
We have to specify how to determine who wins the bet. Now, there’s no big red thermometer sticking out of the North Pole that shows the temperature for the entire Earth. So we have to choose a global temperature “product”. There are several alternatives: GISSTEMP, HadCRUT, RSS, and UAH. The first two use land-based thermometers and the second two use satellite-based microwave sensors. Neither approach is perfect.
However, I refuse to use any of land-based products as a reference for a bet on AGW. They tend to reflect land-use changes as much as climate changes. Two separate papers have concluded that about 40% to 50% of the warming signal from these sources could be attributable to increased economic development around previously rural stations. The actual “boots on the ground” issues with station siting are well documented here. Pave a road or install an air conditioner near a thermometer and voila, instant warming.
Satellite products are no paragon of accuracy either. They actually measure microwave radiation and use a model to infer the temperature. Several refinements over the years have corrected for things like orbital drift, which makes one wonder what other issues are lurking. Moreover, the record contains readings from several different satellites, each with multiple sensors (which degrade over time, BTW). The research groups behind the products use overlapping samples to calibrate new data streams, but that’s not a foolproof process by any means.
In the end, it comes down to coverage. Satellites gives us much denser, homogeneous, and consistent coverage of temperature readings. I think they are therefore much less likely to be systematically biased in detecting trends.
Unfortunately, our quest doesn’t end there. Satellites actually generate temperature data for several different altitudes and latitudes. Which of these best reflect AGW climatic processes? To find out, I consulted with Ross McKitrick and Roy Spencer, two well-known scientists with relevant expertise. He proposes linking AGW-targeted interventions to the tropical troposphere termperature (T3) because all the climate models we currently have produce T3 warming as a unique signature attributable to CO2 increases. Unsurprisingly, Ross asserted that if the counterparty was unwilling to use T3, he didn’t really believe in significant AGW.
I take a slightly different philospohical posture from Ross. The scenarios where T3 is a signature of significant AGW are logically a subset of all scenarios where significant AGW occurs. Sure it could be the same set, but it could be smaller. Therefore, it would not be in the the potential counterparty’s strategic interest to limit the scenarios in which he wins. So I suggested we go with Roy’s suggestion of using the global lower troposphere series as most representative of the climate we care about.
There is one more wrinkle. I had originally suggested we use the three year averages around 2000 and 2020 as the basis for the bet. Given my academic background, I am a bit embarassed by this. Both Ross and Roy pointed out that this bet would be more about big climate events around 2020 than general trends. A big El Nino and I lose. A big volcanic eruption and I win. Not exactly the prediction we intend to measure.
They both suggested we calculate the linear trend from 2000 to 2020 using least squares regression. A decadal trend greater than .10 and I lose. Smaller than .10 and I win. This was the obvious approach in retrospect and I have suggested it to the potential counterparty.
More news as events warrant.